A data-driven approach based health indicator for remaining useful life estimation of bearings

摘要

Accurate state of health monitoring remains an important challenge in the modern industry. Failures in rolling element bearings account for almost half of all machine breakdowns. Consequently, the monitoring of this component is very required for arranging repair work and reducing unexpected production outage. In this sense, this paper presents a new data-driven approach for bearing remaining useful life estimation. Considering the nonlinear and non-stationary bearing vibration signals, the empirical mode decomposition method was used as a pre-processing step for feature extraction. Then, the principal component analysis algorithm was used for feature reduction. Finally, the support vector regression algorithm is proposed to estimate the remaining useful life of the tested bearing. In fact, the proposed method presents a new health indicator that defines accurately the state of health of bearings. Experimental results based on real bearing vibration signals under real operating conditions confirm the reliability and the effectiveness of the proposed method.